QC AI Engineer

Summary:

The Data QC & Model QC Specialist is responsible for ensuring the quality of AI data and Computer Vision / Vision-Language models (VLM) before deploying into real-world products. This role plays a critical part in building a robust, accurate, and scalable datamodel pipeline for AI and Robotics systems.

Requirements

Relevant education and experience

Minimum 3+ years of experience, strong analytical mindset with high attention to data quality and model behavior.

Experience working with AI data pipelines, Computer Vision, VLMs, or multimodal datasets.

Solid understanding of how modern VLMs work (vision encoder + language model, alignment, prompting).

Familiarity with evaluating AI systems beyond raw metrics, focusing on semantic correctness and user-facing behavior.

Experience working with large-scale datasets and structured evaluation workflows.

Good communication skills to collaborate with AI engineers, researchers, and data vendors.

Ability to read and understand technical documentation and research papers in English.

Preferred Qualifications

Hands-on experience with annotation or review tools (e.g. Label Studio, CVAT, or custom review systems).

Basic Python skills for data analysis or evaluation scripting.

Understanding of MLOps / model lifecycle management.

Experience working with external data vendors or crowdsourced annotation teams.

Exposure to robotics, embodied AI, or humanrobot interaction use cases

Personality/ Attitude

Proactive, dedicated, business-oriented, responsible and willing to learn

Good communication skills, creative problem-solving skills and attention to detail

Job description:

Ensure the quality of multimodal datasets (image, video, text, captions, instructions) used for VLM training and evaluation, working with internal teams and external data vendors.

Define and enforce data quality standards for VLMs, covering visual accuracy, textual correctness, visionlanguage alignment, semantic consistency, and bias/noise detection.

Review and audit data before and after annotation, identifying systemic issues such as misaligned captions, hallucinated descriptions, weak prompts, or inconsistent labeling.

Design evaluation datasets and test protocols for VLM use cases including VQA, instruction following, multimodal reasoning, and humanrobot interaction scenarios.

Evaluate VLM performance using both quantitative metrics and qualitative behavior analysis, focusing on correctness, consistency, robustness, latency, and stability in real deployments.

Perform error analysis and deliver clear recommendations for data improvement, prompt refinement, model retraining, and release (go/no-go) decisions.